1. Field of the Invention
This invention relates to polishing of a metal film, and in particular to a method of polishing a metal film in a semiconductor device interconnection process.
2. Description of Related Art
In recent years, as semiconductor integrated circuits (referred to hereafter as LSI) become more complex, new microtechniques are being developed. One of these is chemical mechanical polishing (referred to hereafter as CMP), which is often used in LSI manufacture, in particular flattening of interlayer insulating films, forming metal plugs and inlay of interconnections in multi-layer interconnection processes.
This technology is disclosed, for example, in U.S. Pat. No. 4,944,836.
To achieve higher speeds of LSI, attempts are being made to use low resistance copper alloy instead of the conventional aluminum alloys as an interconnection material, however with copper alloy, microprocessing by dry etching which was used with aluminum alloy is difficult. Therefore the “damascene” method is mainly employed, wherein an inlaid interconnection is formed by depositing a copper alloy thin film on an insulating film on which a groove is formed by dry etching, and the copper alloy thin film is removed by CMP excepting for the part inlaid in the groove. This technique is disclosed for example in Japanese Published Unexamined Patent Application No. 2-278822.
In general, slurries used for CMP of a copper alloy interconnection comprise a solid abrasive and oxidizing substance as the main components. The basic mechanism of CMP is to mechanically remove the oxide by a solid abrasive while oxidizing the surface of the metal by the oxidizing action of an oxidizing substance. This is disclosed on P. 299 of “The Science of CMP”, edited by Masahiro Kashiwagi and published by Science Forum on Aug. 20, 1997 (in Japanese).
As solid abrasives, an alumina abrasive and silica abrasive are known with a particle diameter of a several 10-several 100 nm, but most solid abrasives for metal polishing on the market are of the alumina type.
Generally, as oxidizing substances, hydrogen peroxide (H2O2), ferric nitrate (Fe(NO3)3) and potassium iodate (KlO3) are used, and these are described on p. 299-p. 300 of the aforementioned “Science of CMP”.
However, when interconnections and plugs were formed by CMP using a conventional slurry containing a solid abrasive for metal film polishing as a main component, the following problems (1)-(8) occurred.
(1) Denting (referred to hereafter as dishing) occurs wherein the surface of the central part of the metal interconnection inlaid in the groove formed in the insulating film is polished excessively compared to the periphery, or a phenomenon (referred to hereafter as erosion) occurs wherein the insulating film surface around the interconnection is polished (FIGS. 5A, 5B).
The metal/insulating film selective ratio of a slurry intended for metal film polishing is as high as ten or more. This value is obtained by performing CMP on a wafer with only a flat metal film, and a wafer with only a flat insulating film, and comparing the polishing rates in the two cases.
However, it is known that when CMP is applied to a wafer where a metal film is deposited on an insulating film having a groove which is an interconnection pattern, excessive polishing occurs locally. This is due to the fact that there is unevenness on the surface of the metal film before CMP is performed, reflecting the groove which is the interconnection pattern. When CMP is applied, high pressure occurs locally according to the pattern density, and the polishing rate at these points is faster.
There fore dishing and erosion become conspicuous problems in pads of large area (area of about 0.1 mm side) or with crowded interconnection patterns. These problems are mentioned in J. Electrochem. Soc., p. 2842-2848, Vol. 141, No. 10, October 1994.
(2) Scratches (polishing marks) occur due to the solid abrasive used for polishing. In particular, alumina which is the main material used as a metal polishing abrasive, has a greater hardness than silicon dioxide which is the main material of the insulating film. Therefore, scratches occur on the surface of an insulating film exposed by CMP in addition to the surface of the metal film used for the interconnection. Slurry remains behind in the scratches on the insulating film surface, and causes malfunction of the semiconductor device due to heavy metal ion contamination. It also affects the shape of the upper layer interconnection, and causes short circuits. The scratches on the metal film surface cause poor continuity and deterioration of electromigration resistance.
In order to prevent scratches, down force and platen rotation speed are reduced when CMP is employed. However, it is difficult for even this method to prevent scratches in a soft metal such as copper.
The scratches can be reduced by using a soft polishing pad, but dishing and erosion become more serious, and flatness after CMP deteriorates. It was therefore suggested to perform CMP with a hard polishing pad in a first stage, and then to finish with a soft polishing pad, i.e. to perform a two-stage CMP. A new problem however arises in this case that the throughput falls.
(3) Due to the high frictional force between polishing abrasive and the metal film surface when CMP is performed, peeling occurs between the metal film and the lower insulation layer, or between the spin-on-glass (referred to hereafter as SOG) in the lower insulating layer and the chemical vapor deposition (referred to hereafter as CVD) oxide film. To prevent peeling, the down force and platen rotation speed may be reduced, but if it is attempted to completely prevent it, the CMP rate falls and polishing time becomes longer which is not practical. This can be resolved by using a soft polishing pad, but dishing and erosion become serious, and flatness after CMP deteriorates.(4) As a large amount of polishing abrasive remains behind on the wafer surface after CMP, cleaning must be performed before applying the next step, and foreign matter must be removed until it is below a specified level (e.g., there must be no more than 100 particles of foreign matter greater than 0.2 μm in one wafer). A cleaning machine which employs mechanical cleaning together with chemical cleaning is needed for this purpose.
The cleaning technique is very complicated as shown in FIG. 11. Brush-cleaning and megasonic cleaning that use a reagent fluid together are mainly performed. The brush materials must be special materials which do not damage the metal film surface, and for example, ammonium hydroxide or an aqueous solution of hydrofluoric acid are used as reagent fluid.
Megasonic cleaning is a cleaning method using a high frequency of 800 kHz applied to the cleaning fluid so as to remove abrasive from the substrate. This cleaning is more powerful than conventional cleaning by ultrasonic waves (40 kHz). In this technique, sufficient energy or force must be supplied to remove the abrasive from the substrate. On the other hand the output must be set in a range that does not damage the metal film and insulating film. An example of post-CMP cleaning is disclosed on p. 172 of the May 1995 edition of Semiconductor World (in Japanese).
(5) Consumable items used for CMP are costly. This is because the production cost of abrasives used in the slurry is high, and great care must be taken to adjust particle size. In particular, alumina abrasive is several times higher in price compared with silica abrasive.
In general, foaming polyurethane is used as a polishing pad. When CMP is performed, polishing abrasive adheres to this polishing pad, clogging occurs, and the CMP rate drops.
To prevent this, the polishing pad surface needed to be sharpened with a whetstone (referred to hereafter as a conditioner) to which diamond particles were made to adhere. Therefore the life of the polishing pad was short, and it was a high cost consumable item second to the polishing abrasive. The cost of the CMP process is discussed in “Recent Trends and Problems in CMP Apparatus and Related Materials, Realize Inc., New Tech Lecture, May 1996.
(6) Regarding CMP-related machines and equipment, in addition to the above-mentioned CMP machine and post-cleaning machine, a slurry feeder and processor of waste fluid containing slurry are required. Therefore, the cost of the whole CMP facility became very high. A stirrer is also needed to prevent sedimentation of abrasive in the slurry feeder, and an equipment was required to keep slurry circulating through the piping and stop it depositing. The cost of waste fluid processing is also high, and a recycling technique is needed.(7) It is also a problem that the throughput of the whole CMP process is low. In a CMP facility, it is usual to condition the polishing pad, perform a first CMP to polish the metal film, and perform a second CMP (buff polishing) to remove the damaged layer of insulating film exposed by the first CMP. As the post-cleaning machine involves brush cleaning, wafers are usually cleaned wafer by wafer. Therefore, the throughput of the whole CMP process is the lowest in the semiconductor device manufacturing process. An example of the overall CMP process is given in detail, for example, in the May 1995 edition of Semiconductor World, P. 172.(8) Although the CMP machine uses large amounts of polishing abrasive which generate dust, it must be operated in a clean room. A system must be provided to suppress dust in the exhaust duct of the CMP machine, and a special room must be set up in the clean room to maintain the degree of cleanliness, which is costly.
All the above problems are caused by performing CMP using a slurry containing a highly concentrated polishing abrasive. However, in one prior art CMP method, to increase the polishing rate, the surface of the metal is oxidized by an oxidizer, and the surface of metal that was exposed by mechanically removing this oxide layer with a polishing abrasive is re-oxidized. This process of oxide layer formation and mechanical removal is repeated. In other words, the polishing abrasive was necessary to provide a mechanical removal effect whereby the oxide film could be rapidly removed, and when polishing abrasive was not added, a practical CMP rate was not reached.
In Japanese Published Unexamined Patent Application No. 7-233485, a comparison example is given where CMP was performed with a polishing solution to which polishing abrasive was not added (0.1 wt % aminoacetic acid and 13 wt % hydrogen peroxide). It is reported that in this case, the polishing rate was 10 nm/min, about 1/10 of that of a polishing solution to which an alumina polishing abrasive was added and about 2/7 of that to which a silica polishing abrasive was added.
FIG. 2 is the result of an additional test based on said Japanese Published Unexamined Patent Application No. 7-233485. This measured the hydrogen peroxide aqueous concentration dependency of CMP rate and etching rate in a polishing solution containing 0.1 wt % aminoacetic acid and hydrogen peroxide (not containing abrasive), so as to reproduce the results of the aforesaid Koho. It should be noted that FIG. 2 shows a concentration of 30% aqueous hydrogen peroxide, and to make a comparison with the above Koho, the results should be multiplied by a factor of 0.3. The hard pad IC1000 of the Rodel company was used as a polishing pad. The rotation speeds of the platen (diameter: 340 mm) and holder were both 60 rpm, and the down force was 220 g/cm2 (same as CMP condition of this invention). From the result of FIG. 2, it is seen that when an abrasive is not included, the CMP rate is barely 20 nm/min, i.e. a practical CMP rate is not obtained. When the hydrogen peroxide concentration is low, the etching rate is fast, and stability of polishing becomes poor. The stability rises if the hydrogen peroxide concentration is increased, but the CMP rate becomes very low which is disadvantageous from the viewpoint of throughput.
On further examination, it was also found that the still solution etching rate (the etching rate in the case when a stationary sample was immersed in a polishing solution which was not stirred) does not fall exactly to zero even at high hydrogen peroxide concentration. When the polishing solution is stirred, and the etching rate is measured (the etching rate in a stirred solution is near to the etching rate during CMP), it is seen that the etching rate increases, and exceeds ½ of the polishing rate.
Therefore, it was found that unless the CMP rate was increased by including an abrasive and the ratio of the CMP rate and stirred etching rate (referred to hereafter as rate ratio) was increased, the solution could not be used as a polishing solution. When the rate ratio is low, etching proceeds in depressions not in contact with the polishing surface, and flatness is lost. In fact, using a polishing solution wherein the hydrogen peroxide solution concentration was varied, it was found that a polishing time of from 40 minutes to 1 hour 30 minutes was needed.
A cross-section of the copper interconnection formed is shown in FIGS. 22A, 22B. Most of the copper which would have been left in the groove of the silicon dioxide film was etched out. As a result of a continuity test using a meandering pattern (line width 0.3-3 μm, length 40 mm), the yield was 0%. Therefore, this could not be used as an LSI interconnection. This is due to the fact that as the CMP rate is slow, etching occurred during the long polishing time.
If the concentration of aminoacetic acid is raised, the CMP rate increases, but the stirred etching rate also increases and the same result is obtained. It was found that to suppress etching, potassium hydroxide may be added to the polishing solution to adjust the alkalinity to pH10.5. However, a problem occurs in that the selective ratio falls and erosion occurs due to the etching of the silicon dioxide film by potassium hydroxide. Potassium ion which remains behind spreads through the insulating film, and causes deterioration of the characteristics of the semiconductor device.
This problem is due to the fact that aminoacetic acid itself has not much ability to make copper oxide water-soluble. As seen from the pH-oxidation/reduction potential diagram on p. 387 in M. Pourbaix, “Atlas of Electrochemical Equilibria in Aqueous Solutions”, 1975, published by NACE, and shown in FIG. 9, the range in which copper is made water-soluble as copper ion (domain of corrosion) is pH7 and below, and as aminoacetic acid is neutral, its effect is weak.
FIG. 26 shows the difference of corrosion rate (etching rate) in the domain of corrosion and domain of passivation of copper. The solid line shows the corrosion rate when the oxidation-reduction potential is the same for the citric acid-based polishing solution and the aminoacetic acid-based polishing solution in FIG. 9. As typical examples, corrosion rate was plotted for a polishing solution comprising a mixture of citric acid and aqueous hydrogen peroxide in the domain of corrosion, and a polishing solution comprising a mixture of aminoacetic acid and aqueous hydrogen peroxide in the domain of passivation. Both polishing solutions were prepared with equal mole ratios. Hence, in the domain of corrosion, copper is rendered water-soluble and ionized at a much faster rate than in the domain of passivation.
This is mentioned in Proceedings of the CMP-MIC Conference, 1996, p. 123. Actually, it is reported that aminoacetic acid has no ability to etch copper oxide, but if copper oxide cannot be made water-soluble, it remains on the insulating film which is exposed after performing CMP, and causes electrical short circuits between interconnections. If the slurry contains an abrasive, the copper oxide is easily removed by mechanical action.
Conventional metal etching solutions lie within the above-mentioned domain of corrosion, but it is not certain that they can all be used as CMP polishing solutions for LSI multi-layer interconnections. This is because a slow etching rate is suitable for CMP polishing solutions. This is described, for example, in relation to abrasion experiments on copper surfaces using an aqueous solution of nitric acid, Journal of Abrasive Polishing, p. 231-233, Vol. 41, No. 1, 1997 (in Japanese). It is reported that when there is no abrasive, the CMP rate is low, but due to the absence of scratches, the solution is suitable as a polishing solution. However, the etching rate of this polishing solution was not studied, and there was no attempt to form an interconnection structure. As a result of performing additional tests on this polishing solution, it was found that the still solution etching rate of copper using 1 vol % aqueous nitric acid is 50 nm/min, but a sufficiently large ratio could not be obtained for the CMP rate of 80 nm/min mentioned in the aforesaid Journal. Further, when CMP was applied to form an inlaid interconnection, the copper in the part which should have formed the interconnection was etched and almost completely lost. Hence, polishing can be performed with a polishing solution wherein the etching rate is not suppressed, but an inlaid interconnection cannot be formed.